Automated fundus imaging analysis and diagnostic supporting system

Abstract

The color fundus images which can be obtained noninvasively from individuals plays an important role in the mass screening and diagnosis of eye diseases, especially the diabetic retinopathy. However when there is a large number of images to inspect, diagnose, and annotate, it becomes a burden that demands a considerable amount of time for an ophthalmologist to produce systematically accurate and error free results. Being able to automatically process these image data is therefore very desirable and it has become a challenging research topic that has attracted much attention worldwide. This research project has made a contribution on the development of algorithms to automatically extract fundamental features in the color fundus images. These are the accurate location of optic disk, its boundary estimation, the main courses of blood vessels, and foveal identification. Having successfully obtained these results, a foveal fundus coordinate system is finally set up with exudates detected and localized. In summary, various novel and robust approaches for automatic feature extraction and detection in color fundus images have been proposed. However, the research work is an ongoing process rather than being concluded. Due to the multidisciplinary nature of the problem, closer clinical collaboration and further testing process are necessary.